Development of a Machine Learning Model for Prediction and Parameter Adjustment to Prevent Rotary Kiln Build-up

Abstract

The current investigation focuses on developing a predictive model by applying a machine learning model to the clinker production process, specifically to forecast and adjust operating parameters, reducing the formation of rings and adhesions inside the rotary kiln. To develop the predictive model, a Pearson correlation analysis was performed to identify the operating parameters that contribute most to ring formation, which identified five key variables divided into the categories: chemistry, internal temperature, and kiln feed. Once determined, a decision tree model was used, and operating data was fed into it. The results demonstrated a robust overall accuracy of 91% and an outstanding recall of 97% for the detection of ring formation. This high recall guarantees the minimization of false negatives, which is considered a successful and highly reliable result for early warning, despite the generation of some false alarms (false positives). The integration of the model into an interactive application with an 80% predictive capacity in practice validates its value as a decision-making tool, demonstrating that machine learning is key to optimizing furnace operation.